-
Notifications
You must be signed in to change notification settings - Fork 0
/
features.py
executable file
·1693 lines (1346 loc) · 55.9 KB
/
features.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#!/usr/bin/python
import pronouns as pn
import ucfparser as up
import wordnet as wn
import heads as hd
import saveparses as savep
import simplejson as json
import trees as tr
import operator
import pdb
import math
from nltk.corpus import wordnet as nlwn
import nltk.tree as nltr
import simplejson as json
from glob import glob
import sys
from bz2 import BZ2File
import traceback
import time
from math import log
import shutil
import os
from nltk.corpus.reader.wordnet import information_content
import nltk.corpus.reader.wordnet
import nltk
import nsddebug
import generalizepmi as gpmi
HIGH_SIM = 0.8
LOW_SIM = 0.3
HIGH_FCOUNT = 7.0
relatives_memo = {}
ic_reader = nltk.corpus.reader.wordnet.WordNetICCorpusReader(nltk.data.find('corpora/wordnet_ic'),'.*\.dat')
ic = ic_reader.ic('ic-bnc-resnik.dat')
nsd_debugger = nsddebug.nsd_debugger()
feature_num_to_names = {\
0 : 'subject_head_of_verbs', \
1 : 'object1_head_of_verbs' , \
2 : 'object2_head_of_verbs', \
3 : 'modifiers_of_head', \
4 : 'modified_heads', \
5 : 'prep_head_noun_of_pp_pairs', \
6 : 'prep_head_noun_pairs', \
7 : 'prep_verb_pairs', \
8 : 'coheads'}
name_to_feature_num = {\
'subject_head_of_verbs' : 0, \
'object1_head_of_verbs' : 1 , \
'object2_head_of_verbs' : 2, \
'modifiers_of_head' : 3, \
'modified_heads' : 4, \
'prep_head_noun_of_pp_pairs' : 5, \
'prep_head_noun_pairs' : 6, \
'prep_verb_pairs' : 7, \
'coheads' : 8}
context_names = ['verb_subject_contexts_freqs', \
'verb_object1_contexts_freqs', \
'verb_object2_contexts_freqs', \
'modifier_contexts_freqs', \
'modified_head_contexts_freqs', \
'prep_head_noun_of_pp_pairs_contexts_freqs', \
'prep_head_noun_pairs_contexts_freqs', \
'prep_verb_pairs_contexts_freqs', \
'head_noun_context_freqs']
context_name_to_feature_nums = {\
context_names[0] : 0, \
context_names[1] : 1, \
context_names[2] : 2, \
context_names[3] : 3, \
context_names[4] : 4, \
context_names[5] : 5, \
context_names[6] : 6, \
context_names[7] : 7, \
context_names[8] : 8}
def map_feature_dict_to_names(feature_dict):
new_feature_dict = {}
for val in feature_num_to_names.values():
new_feature_dict[val] = []
for key in feature_dict:
num,feature = key
new_feature_dict[feature_num_to_names[num]].append((feature, feature_dict[key]))
return new_feature_dict
def magnitude(vec):
mag = 0
for v in vec.values():
mag += v*v
return math.sqrt(mag)
def cosine_sim(vec1, vec2):
s1 = set(vec1.keys())
s2 = set(vec2.keys())
common_features = s1.intersection(s2)
dotproduct = 0.0
for feature in common_features:
dotproduct += vec1[feature] * vec2[feature]
#nsd_debugger.write_row( ['%s(<b>%s</b>)' % (feature_num_to_names[feature[0]], feature[1]), vec1[feature]] )
try:
sim = dotproduct / (magnitude(vec1) * magnitude(vec2))
return sim
except ZeroDivisionError: return 0
def pmi(freq_joint, freq_x, freq_y, freq_all):
return log(freq_joint, 2) - log(freq_x, 2) - log(freq_y, 2) + log(freq_all, 2)
class ContextSimilarity(object):
def __init__(self, nounmodels):
self.models = nounmodels
self.context_dict = self.models.context_dict()
def get_pmi_feature_dict(self, stem, feature_dict=None):
if not feature_dict:
stem_fd = self.models.models[stem].feature_dict()
else:
stem_fd = {}
self.models.load_models(None, [stem])
whole_stem_fd = self.models.models[stem].feature_dict()
for k in feature_dict:
if whole_stem_fd.has_key(k):
stem_fd[k] = whole_stem_fd[k]
else:
stem_fd[k] = 1.0
stem_freq = self.models.noun_freqs[stem]
pmi_dict = {}
for feature in stem_fd.keys():
try:
context_freq = self.context_dict[feature]
joint_freq = stem_fd[feature]
pmi_dict[feature] = pmi(joint_freq, stem_freq, context_freq, self.models.word_count)
except KeyError, e:
pass
#print e
except ValueError, e:
pass
#traceback.print_exc(file=sys.stdout)
#print stem, stem_freq, joint_freq, context_freq
return pmi_dict
def similarity_stems(self, stem1, stem2):
stem1_pmi = self.get_pmi_feature_dict(stem1)
stem2_pmi = self.get_pmi_feature_dict(stem2)
return cosine_sim(stem1_pmi, stem2_pmi)
def save_traceback(e):
print 'EXCEPTION', e
f = open('TRACEBACKS', 'a')
traceback.print_exc(file=f)
traceback.print_exc()
f.close()
def modifiers(whole_np_pos_list, head_np_pos_list):
if len(whole_np_pos_list) == len(head_np_pos_list):
return ([], [])
mod = whole_np_pos_list[0:len(whole_np_pos_list) - len(head_np_pos_list)]
# clean_mod words are probably modifiers of the head_np
# dirty_mod words may or may not be modifiers, not recommended to treat them as modifiers
clean_mod = []
dirty_mod = []
mod_list = clean_mod
for m in reversed(mod):
if m.node in ['CC', 'COMMA']:
mod_list = dirty_mod
if m.node not in ['POS', 'CC', 'COMMA', 'SYM', 'FW', 'PERIOD']:
mod_list.insert(0, [m])
return (clean_mod, dirty_mod)
def modifiers_and_head_nouns(tree):
excluded_subtrees = []
word_modifier_pairs = []
for subtree in tree.subtrees():
skip = False
for excluded_tree in excluded_subtrees:
if tr.has_subtree(excluded_tree, subtree):
skip = True
if skip:
continue
if subtree.node in ['NP', 'NP-COORD']:
head_np_list = hd.head_np_of_np(subtree)
for head_np in head_np_list:
excluded_subtrees.append(head_np)
candidates = hd.head_np_candidate_heads(head_np)
if candidates:
head = hd.choose_head_np_from_candidates(candidates)
if head and len(head['STEMS']) > 0:
clean_mod, dirty_mod = modifiers(tr.tree_pos(head_np), head['POS_LIST'])
if len(clean_mod) > 0:
word_modifier_pairs.append( (head['POS_LIST'], clean_mod) )
else:
word_modifier_pairs.append( (head['POS_LIST'], []) )
for dm in dirty_mod:
word_modifier_pairs.append( (dm, []) )
if not isinstance(subtree[0], nltr.Tree):
word_modifier_pairs.append( ([subtree], []) )
return word_modifier_pairs
def pos_list_to_stems(pos_list):
word = ' '.join([pos[0] for pos in pos_list])
if len(pos_list) > 1:
stems = wn.morphy2(word)
if stems:
return ('N', stems)
if len(pos_list) == 1:
pos_word = pos_list[0]
if pos_word.node in ['JJ', 'JJS', 'JJR', 'VBD', 'VBG', 'VBN', 'VBP', 'VBZ', 'VB']:
stems = wn.morphy2(word, pos=nlwn.ADJ)
if stems:
return ('ADJ', stems)
if pos_word.node in ['CD']:
if pos_word[0].startswith('$'):
return ('N', ['monetary_unit'])
try:
if int(pos_word[0]) > 1200 and int(pos_word[0]) < 2200:
return ('N', ['particular_date'])
except ValueError:
pass
return ('N', ['cardinal_number'])
if pos_word.node in ['NN', 'NNS', 'NNP', 'NNPS']:
stems = wn.morphy2(pos_word[0], pos=nlwn.NOUN)
if stems:
return ('N', stems)
if pos_word.node in ['PRP']:
stem = pn.pronoun_mapper.map_to_stem( pos_word[0] )
if stem:
return ('N', [stem])
return ('N', [])
def json_fix_load_dict(j):
j = json.loads(j)
for i in xrange(len(j)):
j[i][0] = tuple(j[i][0])
return tuple([tuple(l) for l in j])
def all_hypernyms(s):
h = {}
for slist in s.hypernym_paths():
for s in slist:
h[s] = True
return h
class PMINounModel(object):
def __init__(self, lemma=None, models_dir=None):
self.noun = None
self.sections = ['subject_head_of_verbs', 'object1_head_of_verbs', \
'object2_head_of_verbs', 'modifiers_of_head', 'modified_heads', \
'prep_head_noun_of_pp_pairs', 'prep_head_noun_pairs', \
'prep_verb_pairs', 'coheads']
for section in self.sections:
self.__dict__[section] = {}
if models_dir and lemma:
self.load_from_pmi_file(models_dir, lemma)
self.fcount = 0
self.high_fcount = 0
def load_from_pmi_file(self, models_dir, lemma):
filename = self.lemma_to_filename(lemma, models_dir)
self._load_from_pmi_file(filename)
def feature_dict(self):
d = {}
for n, fdict_name in enumerate(self.sections):
for k in self.__dict__[fdict_name].keys():
d[ (n, k) ] = self.__dict__[fdict_name][k]
return d
def cosine_similarity_given_other_feature_dict(self, other_model_feature_dict):
self_feature_dict = self.feature_dict()
return cosine_sim(self_feature_dict, other_model_feature_dict)
def lemma_to_filename(self, lemma, dir):
return '%s/%s.pmi.bz2' % (dir, lemma.replace('/', '_'))
def scale(self, weight):
for section in self.sections:
for k in self.__dict__[section]:
self.__dict__[section][k] *= weight
def save_to_file(self, dir):
fout = BZ2File(self.lemma_to_filename(self.noun, dir), 'w')
fout.write('%s\n' % self.noun)
for section in self.sections:
pmi_list = [(k, self.__dict__[section][k]) for k in self.__dict__[section]]
pmi_list.sort(key=operator.itemgetter(1), reverse=True)
save_pmi_list(pmi_list, fout, section)
fout.close()
def intersection(self, other_model, new_name=''):
new_model = PMINounModel(new_name)
for dname in self.sections:
s1 = set(self.__dict__[dname].keys())
s2 = set(other_model.__dict__[dname].keys())
shared_features = s1.intersection(s2)
for dkey in shared_features:
new_model.__dict__[dname][dkey] = min(self.__dict__[dname][dkey], other_model.__dict__[dname][dkey])
return new_model
def union_max(self, pmi_noun_model):
for section in pmi_noun_model.sections:
for k in pmi_noun_model.__dict__[section]:
tup_elem = None
if isinstance(k, tuple):
tup_elem, concept = k
else:
concept = k
if not self.__dict__[section].has_key(k):
self.__dict__[section][k] = pmi_noun_model.__dict__[section][k]
if self.__dict__[section][k] > HIGH_FCOUNT:
self.high_fcount += 1
self.fcount += 1
else:
if self.__dict__[section][k] <= HIGH_FCOUNT and pmi_noun_model.__dict__[section][k] > HIGH_FCOUNT:
self.high_fcount += 1
self.__dict__[section][k] = max(pmi_noun_model.__dict__[section][k], self.__dict__[section][k])
def _load_from_pmi_file(self, filename):
fin = BZ2File(filename, 'r')
self.noun = fin.readline().strip()
self.fcount = 0
self.high_fcount = 0
section = None
for section in self.sections:
self.__dict__[section] = {}
for line in fin:
line = line.strip()
if line.startswith('[') and line.endswith(']'):
section = line[1:-1]
else:
if section == 'coheads' or section == 'modifiers_of_head':
continue
elem, eq, pmi = line.rpartition('=')
elem_parts = tuple(elem.split())
pmi = float(pmi)
if pmi < 3.0: continue
if len(elem_parts) > 2 or len(elem_parts) == 0:
continue
elif len(elem_parts) == 2:
key = elem_parts
elif len(elem_parts) == 1:
key = elem_parts[0]
if pmi > HIGH_FCOUNT:
self.high_fcount += 1
self.fcount += 1
self.__dict__[section][key] = pmi
fin.close()
def unambig_lemma_pairs(lemmas):
for lemma_pair in pairs(lemmas):
s1 = set(nlwn.synsets(lemma_pair[2], pos=nlwn.NOUN))
s2 = set(nlwn.synsets(lemma_pair[3], pos=nlwn.NOUN))
s3 = s1.intersection(s2)
if len(s3) == 1:
yield (lemma_pair[2], lemma_pair[3])
def noisy_concepts(concept, concept_lemma_synsets):
noise = []
for s in concept_lemma_synsets:
if s != concept and s.wup_similarity(concept) < HIGH_SIM:
noise.append(s)
return noise
def synset_is_similar_to_any(synset, synset_list):
for s in synset_list:
if s.wup_similarity(synset) > LOW_SIM:
return True
return False
def hypernyms_by_level(child, level=2):
levels = []
levels.append([child])
for l in xrange(1, level):
if len(levels) == l:
levels.append( [] )
for prev_level_hyp in levels[l-1]:
prev_hyp = prev_level_hyp.hypernyms()
#similar_enough = False
#for h in prev_hyp:
# if h.wup_similarity(child) >= HIGH_SIM:
# similar_enough = True
# break
#if not similar_enough:
# break
levels[l].extend(prev_hyp)
if len(levels[l]) == 0:
break
return levels
def find_intersectable_parent_lemmas(child_synset):
lemma_pairs = []
#parent_lemma_weights = {}
child_hypernym_levels = hypernyms_by_level(child_synset)
for child_lemma in child_synset.lemmas:
child_lemma_synsets = nlwn.synsets(child_lemma.name(), 'n')
if len(child_lemma_synsets) <= 1: continue
noise_synsets = noisy_concepts(child_synset, child_lemma_synsets)
for child_hyp_level in child_hypernym_levels:
for parent_synset in child_hyp_level:
for parent_lemma in parent_synset.lemmas:
if parent_lemma == child_lemma: continue
#print 'parent synset and lemma', parent_lemma
lemma_pair_is_noisy = False
for parent_lemma_synset in nlwn.synsets(parent_lemma.name(), 'n'):
#print '==>', parent_lemma_synset
if parent_synset != parent_lemma_synset and synset_is_similar_to_any(parent_lemma_synset, noise_synsets):
lemma_pair_is_noisy = True
if not lemma_pair_is_noisy:
lemma_pairs.append( (child_lemma.name(), parent_lemma.name()) )
#parent_lemma_weights[parent_lemma.name] = remaining_ic / total_ic
#remaining_ic = information_content(parent_synset, ic)
return set([tuple(sorted(pair)) for pair in lemma_pairs])
def pairs(elem_list):
for i in range(len(elem_list)):
elem1 = elem_list[i]
for offset, elem2 in enumerate(elem_list[i+1:]):
j = offset + i + 1
yield (i, j, elem1, elem2)
class PMISynsetModels(object):
def __init__(self, noun_pmi_models_dir, synset_pmi_models_dir, dest_synset_dir = None):
self.noun_pmi_models_dir = noun_pmi_models_dir
self.synset_pmi_models_dir = synset_pmi_models_dir
self.finished_models = {}
self.finished_pr_models = {}
self.dest_synset_dir = dest_synset_dir
self.models = {}
def load_models(self):
print 'Loading SPMI files'
for filename in glob('%s/*.spmi' % (self.synset_pmi_models_dir)):
model = PMISynsetModel(None, self.noun_pmi_models_dir, self.synset_pmi_models_dir)
model.load_from_synset_pmi_file(filename)
self.models[model.synset_name] = model
print 'done'
def create_mr_pmi_synset_models(self, root_synset=nlwn.synsets('entity', 'n')[0]):
# we have
# 1. current_pmi_synset_model, that's our monosemous relatives from phase I
# 2. hyponym_models, that's the "fully disambiguated" hyponyms
if self.finished_models.has_key(root_synset.name):
pmi_synset_model = PMISynsetModel(root_synset.name, self.noun_pmi_models_dir, self.synset_pmi_models_dir)
pmi_synset_model.load_from_synset_pmi_file()
return pmi_synset_model
hyponym_models = []
for hyponym in root_synset.hyponyms():
hyponym_models.append(self.create_mr_pmi_synset_models(hyponym))
#print '\nAt', root_synset.name
#for h in hyponym_models:
# print '==>', h.synset_name
pmi_synset_model = PMISynsetModel(root_synset.name, self.noun_pmi_models_dir, self.synset_pmi_models_dir)
pmi_synset_model.merge_hyponyms_and_monosemous_lemmas(hyponym_models)
pmi_synset_model.save_to_file()
self.finished_models[pmi_synset_model.synset_name] = True
return pmi_synset_model
def create_pr_pmi_synset_models(self, root_synset=nlwn.synsets('entity', 'n')[0]):
current_pmi_synset_model = PMISynsetModel(root_synset.name, self.noun_pmi_models_dir, self.synset_pmi_models_dir, self.dest_synset_dir)
current_pmi_synset_model.load_from_synset_pmi_file()
if self.finished_pr_models.has_key(root_synset.name):
return current_pmi_synset_model
hyponym_models = []
for hyponym in root_synset.hyponyms():
hyponym_models.append(self.create_pr_pmi_synset_models(hyponym))
intersected_models = []
self.finished_pr_models[current_pmi_synset_model.synset_name] = True
child_synset = nlwn.synset(current_pmi_synset_model.synset_name)
for l1, l2 in find_intersectable_parent_lemmas(child_synset):
if l1 in current_pmi_synset_model.merged_models and l2 in current_pmi_synset_model.merged_models:
continue
try:
l1_noun_model = PMINounModel(l1.lower(), self.noun_pmi_models_dir)
l2_noun_model = PMINounModel(l2.lower(), self.noun_pmi_models_dir)
intersected_models.append(l1_noun_model.intersection(l2_noun_model))
if l1 not in current_pmi_synset_model.merged_models:
current_pmi_synset_model.merged_models.append(l1)
if l2 not in current_pmi_synset_model.merged_models:
current_pmi_synset_model.merged_models.append(l2)
except IOError: pass
for model in hyponym_models:
w = current_pmi_synset_model.weight(model)
current_pmi_synset_model.union_max(model, w)
for model in intersected_models:
current_pmi_synset_model.union_max(model, 1.0)
current_pmi_synset_model.save_to_file()
return current_pmi_synset_model
class PMISynsetModel(object):
def __init__(self, synset_name, noun_pmi_models_dir, synset_pmi_models_dir, dest_synset_dir = None):
self.merged_models = [] # LEMMAS
self.synset_name = synset_name
self.sections = ['subject_head_of_verbs', 'object1_head_of_verbs', \
'object2_head_of_verbs', 'modifiers_of_head', 'modified_heads', \
'prep_head_noun_of_pp_pairs', 'prep_head_noun_pairs', \
'prep_verb_pairs', 'coheads']
for section in self.sections:
self.__dict__[section] = {}
self.noun_pmi_models_dir = noun_pmi_models_dir
self.synset_pmi_models_dir = synset_pmi_models_dir
self.dest_synset_dir = dest_synset_dir
self.ic = information_content(nlwn.synset(self.synset_name), ic)
def cosine_similarity(self, other_model):
self_feature_dict = self.feature_dict()
other_feature_dict = other_model.feature_dict()
return cosine_sim(self_feature_dict, other_feature_dict)
def cosine_similarity_given_other_feature_dict(self, other_model_feature_dict):
print 'cosine sim %s' % self.synset_name
self_feature_dict = self.feature_dict()
return cosine_sim(self_feature_dict, other_model_feature_dict)
def feature_dict(self):
d = {}
for n, fdict_name in enumerate(self.sections):
for k in self.__dict__[fdict_name].keys():
d[ (n, k) ] = self.__dict__[fdict_name][k]
return d
def weight(self, hypo_model):
if hypo_model.ic == 0 or self.ic == 0:
weight = 1.0
else:
weight = self.ic / hypo_model.ic
return weight
# input: loaded synset models for all of our direct hyponyms
# effect: we average_add the hyponyms into us and ...
# look at each of our lemmas:
# if the lemma is monosemous and the lemma is not in self.merged_models,
# load it up from the file, and average_add it
def merge_hyponyms_and_monosemous_lemmas(self, hyponym_models):
for hypo in hyponym_models:
w = self.weight(hypo)
self.union_max(hypo, w)
self.merged_models.extend(hypo.merged_models)
#print '==> merging from hyponym', hypo.synset_name
try:
for lemma in nlwn.synset(self.synset_name).lemmas:
lemma = lemma.name().lower()
try:
if lemma not in self.merged_models and len(nlwn.synsets(lemma, 'n')) == 1:
try:
pmi_noun_model = PMINounModel(lemma, self.noun_pmi_models_dir)
self.union_max(pmi_noun_model, 1.0)
self.merged_models.append(lemma)
#print '==> merging from lemma pmi file', lemma
except IOError, e:
pass
except ValueError, e:
save_traceback(e)
except ValueError, e:
save_traceback(e)
def save_to_file(self):
# write the header:
# synset_name
# lemma1 lemma2 lemma3 ...
if self.dest_synset_dir:
save_dir = self.dest_synset_dir
else:
save_dir = self.synset_pmi_models_dir
fout = open('%s/%s.spmi' % (save_dir, self.synset_name.lower().replace('/', '_')), 'w')
fout.write('%s\n' % self.synset_name)
for m in self.merged_models:
fout.write('%s ' % m)
fout.write('\n')
for section in self.sections:
fout.write('[%s]\n' % section)
sorted_section = sorted([(k, self.__dict__[section][k]) for k in self.__dict__[section]], key=operator.itemgetter(1), reverse=True)
for k, v in sorted_section:
if isinstance(k, tuple):
fout.write('%s %s=%s\n' % (k[0], k[1], v))
else:
fout.write('%s=%s\n' % (k, v))
fout.close()
def load_from_synset_pmi_file(self, filename=None):
if not filename:
fin = open('%s/%s.spmi' % (self.synset_pmi_models_dir, self.synset_name.lower().replace('/', '_')), 'r')
else:
fin = open(filename)
self.synset_name = fin.readline().strip()
for lemma in fin.readline().split():
self.merged_models.append(lemma)
section = None
for line in fin:
line = line.strip()
if line.startswith('[') and line.endswith(']'):
section = line[1:-1]
else:
elem, eq, pmi = line.rpartition('=')
elem_parts = tuple(elem.split())
pmi = float(pmi)
if len(elem_parts) > 2 or len(elem_parts) == 0:
print '???:', line
elif len(elem_parts) == 2:
key = elem_parts
elif len(elem_parts) == 1:
key = elem_parts[0]
self.__dict__[section][key] = pmi
fin.close()
def union_max(self, pmi_noun_model, weight = 1.0):
for section in pmi_noun_model.sections:
for k in pmi_noun_model.__dict__[section]:
if not self.__dict__[section].has_key(k):
self.__dict__[section][k] = pmi_noun_model.__dict__[section][k] * weight
else:
self.__dict__[section][k] = max(pmi_noun_model.__dict__[section][k], self.__dict__[section][k] * weight)
def make_monosemous_relatives(wc_dir, synset_dir):
try:
os.mkdir(synset_dir)
except OSError: pass
models = PMISynsetModels(wc_dir, synset_dir)
models.create_mr_pmi_synset_models()
def make_polysemous_relatives(wc_dir, synset_dir, dest_synset_dir):
try:
os.mkdir(dest_synset_dir)
except OSError: pass
models = PMISynsetModels(wc_dir, synset_dir, dest_synset_dir)
models.create_pr_pmi_synset_models()
def save_pmi_list(pmi_list, fout, feature_name):
fout.write('[%s]\n' % feature_name)
for name, pmi in pmi_list:
#if float(pmi) < 0.01: continue
if isinstance(name, tuple):
fout.write('%s %s=%s\n' % (name[0], name[1], pmi))
else:
fout.write('%s=%s\n' % (name, pmi))
def calculate_and_save_pmi(noun, context_sim, wc_dir):
pmi_dict = context_sim.get_pmi_feature_dict(noun)
pmi_names_dict = {}
for num, feature in pmi_dict:
feature_name = feature_num_to_names[num]
if not pmi_names_dict.has_key(feature_name):
pmi_names_dict[feature_name] = {}
pmi_names_dict[feature_name][feature] = pmi_dict[ (num, feature) ]
fout = BZ2File('%s/%s.pmi.bz2' % (wc_dir, noun.replace('/', '_')), 'w')
fout.write('%s\n' % noun)
for pmi_d_name in pmi_names_dict:
pmi_d = pmi_names_dict[pmi_d_name]
pmi_list = sorted([(name, pmi_d[name]) for name in pmi_d], key=operator.itemgetter(1), reverse=True)
save_pmi_list(pmi_list, fout, pmi_d_name)
fout.close()
class NounModel(object):
#def pretty_print(self):
# fdict = self.feature_dict()
# for f in fdict:
# print f, fdict[f]
def calculate_and_save_pmi(self, noun_models, context_sim, wc_dir):
# global function
calculate_and_save_pmi(self.noun, context_sim, wc_dir)
def save_to_file(self, WC_DIR):
fout = open('%s/%s.freq' % (WC_DIR, self.noun.replace('/', '_')), 'w')
fout.write(self.save_as_string())
fout.close()
def feature_dict(self):
feature_list = [self.subject_head_of_verbs, self.object1_head_of_verbs, self.object2_head_of_verbs, \
self.modifiers_of_head, self.modified_heads, self.prep_head_noun_of_pp_pairs, \
self.prep_head_noun_pairs, self.prep_verb_pairs, self.coheads]
d = {}
for n, fdict in enumerate(feature_list):
for k in fdict.keys():
d[ (n, k) ] = fdict[k]
return d
def merge(self, other_model):
for dname in self.feature_dict_names:
for k in other_model.__dict__[dname]:
if not self.__dict__[dname].has_key(k):
self.__dict__[dname][k] = 0
self.__dict__[dname][k] += other_model.__dict__[dname][k]
def __init__(self, noun=''):
self.noun = noun
self.feature_dict_names = ['subject_head_of_verbs', 'object1_head_of_verbs', \
'object2_head_of_verbs', 'modifiers_of_head', 'modified_heads', \
'prep_head_noun_of_pp_pairs', 'prep_head_noun_pairs', \
'prep_verb_pairs', 'coheads']
self.sections = self.feature_dict_names
# when noun acts as the head of a subject, then subject_head_of_verbs
# are those verbs -- lowercase and lemmatized
#
# what verbs tend to select for noun?
self.subject_head_of_verbs = {}
self.object1_head_of_verbs = {}
self.object2_head_of_verbs = {}
# when noun acts as a head of an NP, then modifiers_of_head are the
# words that come before it within the NP -- lowercase and lemmatized
#
# what modifiers tend to select for noun?
self.modifiers_of_head = {}
self.modified_heads = {}
# when noun acts as a head, these are the
# (preposition heading the PP, head noun of the NP embedded within the PP)
# pairs, when a PP may be attached to the noun (often times it is
# attached to the verb, and it is added anyways -- but if a noun truly
# selects for a preposition, there should be a good sign in here)
#
# what prepositions and head nouns of those PPs does noun tend to select for?
self.prep_head_noun_of_pp_pairs = {}
# when noun is head of a PP, these are the (prep, noun_head) pairs
# where noun_head is what the PP attaches to
self.prep_head_noun_pairs = {}
# when noun is head of a PP, these are the (prep, verb) pairs
# where verb is what the PP attaches to
self.prep_verb_pairs = {}
self.coheads = {}
self.coheads_synsets = {}
self.prep_head_noun_pairs_synsets = {}
self.prep_head_noun_of_pp_pairs_synsets = {}
self.modified_heads_synsets = {}
self.coheads_scores = []
self.prep_noun_scores = []
self.prep_noun_pp_scores = []
self.modified_heads_scores = []
self.last_wc_dir = None
def save_as_string(self):
subject = json.dumps(tuple([(k, self.subject_head_of_verbs[k]) for k in self.subject_head_of_verbs]))
object1 = json.dumps(tuple([(k, self.object1_head_of_verbs[k]) for k in self.object1_head_of_verbs]))
object2 = json.dumps(tuple([(k, self.object2_head_of_verbs[k]) for k in self.object2_head_of_verbs]))
modifiers = json.dumps(tuple([(k, self.modifiers_of_head[k]) for k in self.modifiers_of_head]))
modifieds = json.dumps(tuple([(k, self.modified_heads[k]) for k in self.modified_heads]))
pp = json.dumps(tuple([(k, self.prep_head_noun_pairs[k]) for k in self.prep_head_noun_pairs]))
pp2 = json.dumps(tuple([(k, self.prep_head_noun_of_pp_pairs[k]) for k in self.prep_head_noun_of_pp_pairs]))
pp3 = json.dumps(tuple([(k, self.prep_verb_pairs[k]) for k in self.prep_verb_pairs]))
coheads = json.dumps(tuple([(k, self.coheads[k]) for k in self.coheads]))
return json.dumps((self.noun, subject, object1, object2, modifiers, modifieds, pp, pp2, pp3, coheads))
def load_from_string(self, json_str):
self.noun, subject, object1, object2, modifiers, modifieds, pp, pp2, pp3, coheads = json.loads(json_str)
subject = tuple([tuple(l) for l in json.loads(subject)])
object1 = tuple([tuple(l) for l in json.loads(object1)])
object2 = tuple([tuple(l) for l in json.loads(object2)])
modifieds = tuple([tuple(l) for l in json.loads(modifieds)])
coheads = tuple([tuple(l) for l in json.loads(coheads)])
pp = json_fix_load_dict(pp)
pp2 = json_fix_load_dict(pp2)
pp3 = json_fix_load_dict(pp3)
modifiers = json_fix_load_dict(modifiers)
self.subject_head_of_verbs = dict(subject)
self.object1_head_of_verbs = dict(object1)
self.object2_head_of_verbs = dict(object2)
self.modifiers_of_head = dict(modifiers)
self.modified_heads = dict(modifieds)
self.prep_head_noun_pairs = dict(pp)
self.prep_head_noun_of_pp_pairs = dict(pp2)
self.prep_verb_pairs = dict(pp3)
self.coheads = dict(coheads)
def pretty_print(self, min=0):
print 'MODIFIERS of << %s >> when it acts as a head of an NP' % self.noun
print '%-20s%s' % ('Modifier','Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.modifiers_of_head[k]) for k in self.modifiers_of_head], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%.0f' % (k,v)
print 'HEADS MODIFIED by << %s >> ' % self.noun
print '%-20s%s' % ('Head','Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.modified_heads[k]) for k in self.modified_heads], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%.0f' % (k,v)
print 'CO-HEADS of << %s >>' % self.noun
print '%-20s%s' % ('Co-head','Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.coheads[k]) for k in self.coheads], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%.0f' % (k,v)
print '\nVERBS of scopes in which << %s >> acts as the SUBJECT\'s head noun' % self.noun
print '%-20s%s' % ('Verb','Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.subject_head_of_verbs[k]) for k in self.subject_head_of_verbs], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%.1f' % (k,v)
print '\nVERBS of scopes in which << %s >> acts as the OBJECT-1\'s head noun' % self.noun
print '%-20s%s' % ('Verb','Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.object1_head_of_verbs[k]) for k in self.object1_head_of_verbs], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%.1f' % (k,v)
print '\nVERBS of scopes in which << %s >> acts as the OBJECT-2\'s head noun' % self.noun
print '%-20s%s' % ('Verb','Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.object2_head_of_verbs[k]) for k in self.object2_head_of_verbs], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%.1f' % (k,v)
print '\n(PREP, HEAD-NOUN-OF-NP-EMBEDDED-WITHIN-PP) pairs which possibly attach to << %s >>' % self.noun
print '%-20s%-20s%s' % ('Preposition', 'Head PP', 'Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.prep_head_noun_of_pp_pairs[k]) for k in self.prep_head_noun_of_pp_pairs], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%-20s%.1f' % (k[0], k[1], v)
print '\n(PREP, HEAD-NOUN) pairs to which PP headed by << %s >> attaches' % self.noun
print '%-20s%-20s%s' % ('Preposition', 'Head Noun of NP', 'Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.prep_head_noun_pairs[k]) for k in self.prep_head_noun_pairs], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%-20s%.1f' % (k[0], k[1], v)
print '\n(PREP, VERB) pairs to which PP headed by << %s >> attaches' % self.noun
print '%-20s%-20s%s' % ('Preposition', 'Head Noun of NP', 'Frequency')
print '-----------------------------------------------------'
m = sorted([(k, self.prep_verb_pairs[k]) for k in self.prep_verb_pairs], key=operator.itemgetter(1), reverse=True)
for k,v in m:
if v > min:
print '%-20s%-20s%.1f' % (k[0], k[1], v)
def add_subject_head_of_verb(self, verb, credit):
if not self.subject_head_of_verbs.has_key(verb):
self.subject_head_of_verbs[verb] = 0
self.subject_head_of_verbs[verb] += credit
def add_object1_head_of_verb(self, verb, credit):
if not self.object1_head_of_verbs.has_key(verb):
self.object1_head_of_verbs[verb] = 0
self.object1_head_of_verbs[verb] += credit
def add_object2_head_of_verb(self, verb, credit):
if not self.object2_head_of_verbs.has_key(verb):
self.object2_head_of_verbs[verb] = 0
self.object2_head_of_verbs[verb] += credit
def add_modifier_of_head(self, modifier, credit):
if not self.modifiers_of_head.has_key(modifier):
self.modifiers_of_head[modifier] = 0
self.modifiers_of_head[modifier] += credit
def add_modified_head(self, head, credit):
if not self.modified_heads.has_key(head):
self.modified_heads[head] = 0
self.modified_heads[head] += credit
def add_cohead(self, cohead, credit):
if not self.coheads.has_key(cohead):
self.coheads[cohead] = 0
self.coheads[cohead] += credit
def add_prep_head_noun_of_pp_pair(self, prep, head_pp, credit):
if not self.prep_head_noun_of_pp_pairs.has_key((prep, head_pp)):
self.prep_head_noun_of_pp_pairs[(prep, head_pp)] = 0
self.prep_head_noun_of_pp_pairs[(prep, head_pp)] += credit
def add_prep_head_noun_pair(self, prep, head_noun, credit):
if not self.prep_head_noun_pairs.has_key((prep, head_noun)):